Texture Image Classification Using Effective Texture Descriptors

Image texture refers to the visual patterns, variations, or configurations of pixel intensities within an image. Classifying textures is a fundamental goal in computer vision, applicable in areas ranging from medical picture analysis to distant sensing. Throughout the years, numerous strategies have been proposed to address this challenge; however, recent advances in deep learning have significantly transformed the subject. The proposed work delineates reliable and resilient local descriptors termed Texture Classification using Effective Texture Descriptors (TCETD), which integrates Locally Directional and Extremal Pattern (LDEP) with Gray-Level Co-occurrence Matrix (GLCM) to effectively acquire directional, extremum statistics, and spatial relationships among pixel intensities. To communicate directions related to the local area, it first obtains the directional local difference count pattern (DLDCP), which is divided into symmetric and asymmetric positions. By integrating the extremum location, differential, and compression pattern from adjacent sites, we extract the neighbor's extremum-related local pattern to acquire the extremum data generated by the initial segment. The two elements are combined to create the LDEP. The GLCM extracts spatial correlations, pixel intensity patterns, and features based on the distance and angle of pixels within an image. This descriptor can be utilized alongside the LDEP approach to offer a more thorough and resilient representation of the picture texture features, hence enhancing classification accuracy. The outcomes of experiments performed on three notable texture image databases-Klyberg (Stex), Kth-tips2-a, and CUReT-exhibit comparable correct classification rates of 97.91%, 93.82%, and 97.25%, respectively. These rates were achieved using our recently proposed TCETD descriptor under diverse conditions, including rotational and illumination variations, scale differences, and viewpoint alterations, in contrast to traditional methods for classifying texture images. The efficacy of the proposed strategy is corroborated using the Bonn BTF dataset, and the recommended method demonstrated superior performance.